The 28 September 2004 Parkfield, California, earthquake (Mw 6.0) and four aftershocks (Mw 4.7-5.1) were recorded on 12 accelerograph stations of the U.S. Geological Survey Parkfield seismic array (UPSAR), an array of three-component accelerographs occupying an area of about 1 km2 located 8.8 km from the San Andreas fault. Peak horizontal acceleration and velocity at UPSAR during the mainshock were 0.45g and 27 cm/sec, respectively. We determined both time-varying and peak values of ground dilatations, shear strains, torsions, tilts, torsion rates, and tilt rates by applying a time-dependent geodetic analysis to the observed array displacement time series. Array-derived dilatations agree fairly well with point measurements made on high sample rate recordings of the Parkfield-area dilatometers (Johnston et al., 2006). Torsion Fourier amplitude spectra agree well with ground velocity spectra, as expected for propagating plane waves. A simple predictive relation, using the predicted peak velocity from the Boore-Atkinson ground-motion prediction relation (Boore and Atkinson, 2007) scaled by a phase velocity of 1 km/sec, predicts observed peak Parkfield and Chi-Chi rotations (Huang, 2003) well. However, rotation rates measured during Mw 5 Ito, Japan, events observed on a gyro sensor (Takeo, 1998) are factors of 5-60 greater than those predicted by our predictive relation. This discrepancy might be caused by a scale dependence in rotation, with rotations measured over a short baseline exceeding those measured over long baselines. An alternative hypothesis is that events having significant non-double-couple mechanisms, like the Ito events, radiate much stronger rotations than double-couple events. If this is true, then rotational observations might provide an important source of new information for monitoring seismicity in volcanic areas.